Related Experiment Video
Updated: Jan 10, 2026
06:29
Wireless Telemetry Device Implantation in a Fontan Ovine Model for Continuous and Long-Term Hemodynamic Monitoring
Published on: May 2, 2025
686
Prediction of Postoperative Mortality After Fontan Procedure: A Clinical Prediction Model Study Using Deep Learning
Jacek Kolcz1, Anna Budzynska1, Justyna Stefaniak2
1Department of Pediatric Cardiac Surgery, Collegium Medicum, Jagiellonian University, Wielicka 265 St., 31-007 Krakow, Poland.
Journal of Cardiovascular Development and Disease
|November 26, 2025
Summary
A deep learning model accurately predicts postoperative mortality after Fontan surgery, improving risk stratification for single-ventricle congenital heart disease patients. Key factors like pulmonary artery pressure are identified, enhancing personalized care.
Area of Science:
- Cardiology
- Medical Artificial Intelligence
- Computational Biology
Background:
- The Fontan procedure is a critical surgery for single-ventricle congenital heart disease (CHD).
- Postoperative and long-term risks associated with the Fontan procedure necessitate improved risk stratification methods.
- Current risk models for Fontan surgery have limitations in providing accurate, individualized predictions.
Purpose of the Study:
- To develop and validate a deep learning (DL) model for predicting postoperative mortality after Fontan procedure.
- To identify key factors influencing mortality risk in Fontan surgery patients.
- To create a user-friendly tool for personalized risk assessment.
Main Methods:
- Retrospective analysis of 230 patients undergoing Fontan procedure (2010-2024).
- Development of a Deep Neural Network (DNN) model using comprehensive clinical, biochemical, and hemodynamic data.
- Utilized five-fold cross-validation, SMOTE for class imbalance, and SHAP for interpretability, with a Streamlit interface for clinical application.
Main Results:
- The DNN model achieved high predictive performance: 91.5% accuracy, 83.3% precision, 90.9% recall, and 0.94 AUC-ROC.
- SHAP analysis identified pulmonary artery pressure, ventricular end-diastolic pressure, BNP levels, and AV valve regurgitation severity as key mortality predictors.
- A Streamlit application was developed for accessible, personalized risk evaluation.
Conclusions:
- A DL model utilizing detailed clinical data can accurately predict postoperative mortality in Fontan surgery.
- AI-driven risk assessment, enhanced by interpretability, offers a valuable tool for personalized patient care.
- This approach has the potential to improve preoperative counseling, perioperative management, and patient outcomes, pending external validation.
